Chinese startup DeepSeek is making its AI model code publicly available, marking a significant move towards open-source artificial intelligence. The company will release five code repositories next week, describing the action as “small but sincere progress” that will be shared with full transparency.
DeepSeek gained international attention last month when it released its open-source R1 reasoning model, rivaling Western systems in performance while being developed at a lower cost. This move has set the company apart from most AI firms in China, which typically favor closed-sourced models. Founder Liang Wenfeng stated that the firm’s commitment to open-source is driven by soft power and a desire for respect within the community.
The newly released code will provide infrastructure support for existing open source model frameworks, building on top of those already shared. This move comes after DeepSeek recently launched a new algorithm called Native Sparse Attention (NSA), designed to improve long-context training and inference efficiency.
DeepSeek’s user base has surged since last month, with 22.2 million daily active users in China, surpassing Douban’s 16.95 million users as of January 11. The company’s decision to open-source its AI models is seen as a significant step towards the global industry shift towards more open and collaborative approaches.
Source: https://www.reuters.com/technology/artificial-intelligence/deepseek-share-some-ai-model-code-doubling-down-open-source-2025-02-21